Andreessen Horowitz wagers $3 billion to avoid the AI bubble

Investor enthusiasm for artificial intelligence has spilled rapidly into the once staid world of developer tools, benchmarking platforms and backend software, pushing valuations to record highs and fuelling a new wave of technology investment led by Andreessen Horowitz.

An AI startup that helps developers write and debug code is now valued at nearly the same level as United Airlines, while a two-month-old AI computer company has raised an unprecedented $475 million seed round and plans to raise more capital soon. Meanwhile, an AI model-ranking platform has reached a valuation close to $2 billion less than a year after being spun out of an academic project.

At the centre of many of these deals is Andreessen Horowitz, or a16z, which launched a dedicated $1.25 billion fund in 2024 to back AI infrastructure companies. Earlier this month, the Silicon Valley venture capital firm said it would commit an additional $1.7 billion to the strategy.

Unlike traditional definitions that focus on chips and data centres, a16z defines AI infrastructure broadly to include software sold to technical buyers, such as coding tools, foundational AI models, networking and security platforms. “Some of the most important companies of tomorrow will be infrastructure companies,” said Raghu Raghuram, managing partner at a16z and former VMware chief executive.

The firm’s bets are already yielding high-profile outcomes. In recent months, Stripe agreed to acquire Andreessen-backed billing startup Metronome for a reported $1 billion, Salesforce bought Regrello, an AI software provider for manufacturers, and Meta Platforms acquired AI audio company WaveForms. In November, AI coding startup Cursor raised a new funding round at a valuation of $29.3 billion, sharply higher than its $400 million valuation in 2024, when a16z first invested.

Ben Horowitz, co-founder of Andreessen Horowitz, cautioned that it remains too early to judge the fund’s long-term performance, which is typically assessed over a decade. “But so far,” he said, “it’s one of the best funds I’ve ever seen.”

Still, the scale and speed of valuation growth has reignited concerns about a potential AI bubble. A key question for investors is whether enterprises will ultimately spend enough on AI software to justify the trillions of dollars now flowing into the sector.

Steering a16z’s AI infrastructure strategy is Martin Casado, a former computational physicist and entrepreneur who sold his startup Nicira to VMware for $1.26 billion. Casado joined a16z a decade ago and is now seen as a successor to Horowitz in infrastructure investing. While acknowledging that “private valuations are crazy,” Casado said he is not worried about an AI bubble, pointing to strong customer demand and rising usage of computing resources.

The firm has been selective, avoiding direct investments in large-scale AI data centres, though Casado has said he regrets passing on so-called “neocloud” providers such as CoreWeave, now valued at about $50 billion. Typically, the fund writes relatively small early-stage cheques, with its largest investments around $60 million.

Andreessen Horowitz’s approach relies heavily on a technically deep team with unconventional backgrounds. Casado, who grew up in modest circumstances across Europe and the US, surrounds himself with former engineers and product leaders, including Jennifer Li and Guido Appenzeller, to identify promising startups early.

That strategy helped the firm spot Cursor ahead of rivals and back projects such as World Labs, founded by AI pioneer Fei-Fei Li, which focuses on building “world models” to simulate physical environments. The firm is also known for its hands-on involvement, assisting portfolio companies with hiring, infrastructure challenges and, when necessary, exit negotiations.

Casado has played behind-the-scenes roles in several acquisitions, including Cloudflare’s purchase of Replicate and Databricks’ acquisition of Tabular, helping steer outcomes favourable to Andreessen-backed startups.

Despite the current boom, Casado remains realistic about the risks ahead. “There’s going to be a bunch of companies that don’t work,” he said. “There are always fewer winners than people assume, but they end up being much larger than people expect.”